Early Prediction of Alzheimer’s Disease with a Multimodal Multitask Deep Learning Model

Abstract

Alzheimer’s disease (AD) is the sixth leading cause of death in the United States and the most common neurodegenerative disease in adults over 65. Early-stage AD is often misinterpreted as normal cognitive aging because it may not cause adverse symptoms or visible behavioral changes for up to 20 years. Machine learning has been used to avoid misinterpretation of data and more accurately predict the onset of AD. This study aims to use the data typically available in a clinical setting to predict the onset of AD while maintaining a high level of accuracy. This study proposes a deep learning model that uses multimodal input data and performs multitask classification to predict AD diagnosis and scores of two commonly used cognitive assessments: Alzheimer’s Disease Assessment Scale (ADAS) and Mini-Mental State Examination (MMSE). The model was validated using the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset of 1737 patients. The current model achieved a greater accuracy in predicting AD diagnosis and a lower error in predicting ADAS and MMSE scores than existing state-of-the-art models. This model can be applied to the clinical setting so that accurate diagnosis can be achieved, and appropriate action can be taken. Future investigations could include using a convolutional neural network (CNN) to process data from clinical images directly or training and validating the model with other clinical datasets to further improve its accuracy.